includes/normal/UtfNormal.php

   1 <?php
   2 # Copyright (C) 2004 Brion Vibber <brion@pobox.com>
   3 # http://www.mediawiki.org/
   4 #
   5 # This program is free software; you can redistribute it and/or modify
   6 # it under the terms of the GNU General Public License as published by
   7 # the Free Software Foundation; either version 2 of the License, or
   8 # (at your option) any later version.
   9 #
  10 # This program is distributed in the hope that it will be useful,
  11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  13 # GNU General Public License for more details.
  14 #
  15 # You should have received a copy of the GNU General Public License along
  16 # with this program; if not, write to the Free Software Foundation, Inc.,
  17 # 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18 # http://www.gnu.org/copyleft/gpl.html
  19
  20 /**
  21  * Unicode normalization routines for working with UTF-8 strings.
  22  * Currently assumes that input strings are valid UTF-8!
  23  *
  24  * Not as fast as I'd like, but should be usable for most purposes.
  25  * UtfNormal::toNFC() will bail early if given ASCII text or text
  26  * it can quickly deterimine is already normalized.
  27  *
  28  * All functions can be called static.
  29  *
  30  * See description of forms at http://www.unicode.org/reports/tr15/
  31  *
  32  * @package UtfNormal
  33  */
  34
  35 /** */
  36 require_once 'UtfNormalUtil.php';
  37
  38 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
  39 $utfCombiningClass = NULL;
  40 $utfCanonicalComp = NULL;
  41 $utfCanonicalDecomp = NULL;
  42
  43 # Load compatibility decompositions on demand if they are needed.
  44 global $utfCompatibilityDecomp;
  45 $utfCompatibilityDecomp = NULL;
  46
  47 define( 'UNICODE_HANGUL_FIRST', 0xac00 );
  48 define( 'UNICODE_HANGUL_LAST',  0xd7a3 );
  49
  50 define( 'UNICODE_HANGUL_LBASE', 0x1100 );
  51 define( 'UNICODE_HANGUL_VBASE', 0x1161 );
  52 define( 'UNICODE_HANGUL_TBASE', 0x11a7 );
  53
  54 define( 'UNICODE_HANGUL_LCOUNT', 19 );
  55 define( 'UNICODE_HANGUL_VCOUNT', 21 );
  56 define( 'UNICODE_HANGUL_TCOUNT', 28 );
  57 define( 'UNICODE_HANGUL_NCOUNT', UNICODE_HANGUL_VCOUNT * UNICODE_HANGUL_TCOUNT );
  58
  59 define( 'UNICODE_HANGUL_LEND', UNICODE_HANGUL_LBASE + UNICODE_HANGUL_LCOUNT - 1 );
  60 define( 'UNICODE_HANGUL_VEND', UNICODE_HANGUL_VBASE + UNICODE_HANGUL_VCOUNT - 1 );
  61 define( 'UNICODE_HANGUL_TEND', UNICODE_HANGUL_TBASE + UNICODE_HANGUL_TCOUNT - 1 );
  62
  63 define( 'UNICODE_SURROGATE_FIRST', 0xd800 );
  64 define( 'UNICODE_SURROGATE_LAST', 0xdfff );
  65 define( 'UNICODE_MAX', 0x10ffff );
  66 define( 'UNICODE_REPLACEMENT', 0xfffd );
  67
  68
  69 define( 'UTF8_HANGUL_FIRST', codepointToUtf8( UNICODE_HANGUL_FIRST ) );
  70 define( 'UTF8_HANGUL_LAST', codepointToUtf8( UNICODE_HANGUL_LAST ) );
  71
  72 define( 'UTF8_HANGUL_LBASE', codepointToUtf8( UNICODE_HANGUL_LBASE ) );
  73 define( 'UTF8_HANGUL_VBASE', codepointToUtf8( UNICODE_HANGUL_VBASE ) );
  74 define( 'UTF8_HANGUL_TBASE', codepointToUtf8( UNICODE_HANGUL_TBASE ) );
  75
  76 define( 'UTF8_HANGUL_LEND', codepointToUtf8( UNICODE_HANGUL_LEND ) );
  77 define( 'UTF8_HANGUL_VEND', codepointToUtf8( UNICODE_HANGUL_VEND ) );
  78 define( 'UTF8_HANGUL_TEND', codepointToUtf8( UNICODE_HANGUL_TEND ) );
  79
  80 define( 'UTF8_SURROGATE_FIRST', codepointToUtf8( UNICODE_SURROGATE_FIRST ) );
  81 define( 'UTF8_SURROGATE_LAST', codepointToUtf8( UNICODE_SURROGATE_LAST ) );
  82 define( 'UTF8_MAX', codepointToUtf8( UNICODE_MAX ) );
  83 define( 'UTF8_REPLACEMENT', codepointToUtf8( UNICODE_REPLACEMENT ) );
  84 #define( 'UTF8_REPLACEMENT', '!' );
  85
  86 define( 'UTF8_OVERLONG_A', "\xc1\xbf" );
  87 define( 'UTF8_OVERLONG_B', "\xe0\x9f\xbf" );
  88 define( 'UTF8_OVERLONG_C', "\xf0\x8f\xbf\xbf" );
  89
  90 # These two ranges are illegal
  91 define( 'UTF8_FDD0', codepointToUtf8( 0xfdd0 ) );
  92 define( 'UTF8_FDEF', codepointToUtf8( 0xfdef ) );
  93 define( 'UTF8_FFFE', codepointToUtf8( 0xfffe ) );
  94 define( 'UTF8_FFFF', codepointToUtf8( 0xffff ) );
  95
  96 define( 'UTF8_HEAD', false );
  97 define( 'UTF8_TAIL', true );
  98
  99
 100 /**
 101  * For using the ICU wrapper
 102  */
 103 define( 'UNORM_NONE', 1 );
 104 define( 'UNORM_NFD',  2 );
 105 define( 'UNORM_NFKD', 3 );
 106 define( 'UNORM_NFC',  4 );
 107 define( 'UNORM_DEFAULT', UNORM_NFC );
 108 define( 'UNORM_NFKC', 5 );
 109 define( 'UNORM_FCD',  6 );
 110
 111 define( 'NORMALIZE_ICU', function_exists( 'utf8_normalize' ) );
 112
 113 /**
 114  *
 115  * @package MediaWiki
 116  */
 117 class UtfNormal {
 118         /**
 119          * The ultimate convenience function! Clean up invalid UTF-8 sequences,
 120          * and convert to normal form C, canonical composition.
 121          *
 122          * Fast return for pure ASCII strings; some lesser optimizations for
 123          * strings containing only known-good characters. Not as fast as toNFC().
 124          *
 125          * @param string $string a UTF-8 string
 126          * @return string a clean, shiny, normalized UTF-8 string
 127          */
 128         function cleanUp( $string ) {
 129                 if( NORMALIZE_ICU ) {
 130                         # We exclude a few chars that ICU would not.
 131                         $string = preg_replace(
 132                                 '/[\x00-\x08\x0b\x0c\x0e-\x1f]/',
 133                                 UTF8_REPLACEMENT,
 134                                 $string );
 135                         $string = str_replace( UTF8_FFFE, UTF8_REPLACEMENT, $string );
 136                         $string = str_replace( UTF8_FFFF, UTF8_REPLACEMENT, $string );
 137
 138                         # UnicodeString constructor fails if the string ends with a
 139                         # head byte. Add a junk char at the end, we'll strip it off.
 140                         return rtrim( utf8_normalize( $string . "\x01", UNORM_NFC ), "\x01" );
 141                 } elseif( UtfNormal::quickIsNFCVerify( $string ) ) {
 142                         # Side effect -- $string has had UTF-8 errors cleaned up.
 143                         return $string;
 144                 } else {
 145                         return UtfNormal::NFC( $string );
 146                 }
 147         }
 148
 149         /**
 150          * Convert a UTF-8 string to normal form C, canonical composition.
 151          * Fast return for pure ASCII strings; some lesser optimizations for
 152          * strings containing only known-good characters.
 153          *
 154          * @param string $string a valid UTF-8 string. Input is not validated.
 155          * @return string a UTF-8 string in normal form C
 156          */
 157         function toNFC( $string ) {
 158                 if( NORMALIZE_ICU )
 159                         return utf8_normalize( $string, UNORM_NFC );
 160                 elseif( UtfNormal::quickIsNFC( $string ) )
 161                         return $string;
 162                 else
 163                         return UtfNormal::NFC( $string );
 164         }
 165
 166         /**
 167          * Convert a UTF-8 string to normal form D, canonical decomposition.
 168          * Fast return for pure ASCII strings.
 169          *
 170          * @param string $string a valid UTF-8 string. Input is not validated.
 171          * @return string a UTF-8 string in normal form D
 172          */
 173         function toNFD( $string ) {
 174                 if( NORMALIZE_ICU )
 175                         return utf8_normalize( $string, UNORM_NFD );
 176                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 177                         return UtfNormal::NFD( $string );
 178                 else
 179                         return $string;
 180         }
 181
 182         /**
 183          * Convert a UTF-8 string to normal form KC, compatibility composition.
 184          * This may cause irreversible information loss, use judiciously.
 185          * Fast return for pure ASCII strings.
 186          *
 187          * @param string $string a valid UTF-8 string. Input is not validated.
 188          * @return string a UTF-8 string in normal form KC
 189          */
 190         function toNFKC( $string ) {
 191                 if( NORMALIZE_ICU )
 192                         return utf8_normalize( $string, UNORM_NFKC );
 193                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 194                         return UtfNormal::NFKC( $string );
 195                 else
 196                         return $string;
 197         }
 198
 199         /**
 200          * Convert a UTF-8 string to normal form KD, compatibility decomposition.
 201          * This may cause irreversible information loss, use judiciously.
 202          * Fast return for pure ASCII strings.
 203          *
 204          * @param string $string a valid UTF-8 string. Input is not validated.
 205          * @return string a UTF-8 string in normal form KD
 206          */
 207         function toNFKD( $string ) {
 208                 if( NORMALIZE_ICU )
 209                         return utf8_normalize( $string, UNORM_NFKD );
 210                 elseif( preg_match( '/[\x80-\xff]/', $string ) )
 211                         return UtfNormal::NFKD( $string );
 212                 else
 213                         return $string;
 214         }
 215
 216         /**
 217          * Load the basic composition data if necessary
 218          * @access private
 219          */
 220         function loadData() {
 221                 global $utfCombiningClass, $utfCanonicalComp, $utfCanonicalDecomp;
 222                 if( !isset( $utfCombiningClass ) ) {
 223                         require_once( 'UtfNormalData.inc' );
 224                 }
 225         }
 226
 227         /**
 228          * Returns true if the string is _definitely_ in NFC.
 229          * Returns false if not or uncertain.
 230          * @param string $string a valid UTF-8 string. Input is not validated.
 231          * @return bool
 232          */
 233         function quickIsNFC( $string ) {
 234                 # ASCII is always valid NFC!
 235                 # If it's pure ASCII, let it through.
 236                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 237
 238                 UtfNormal::loadData();
 239                 global $utfCheckNFC, $utfCombiningClass;
 240                 $len = strlen( $string );
 241                 for( $i = 0; $i < $len; $i++ ) {
 242                         $c = $string{$i};
 243                         $n = ord( $c );
 244                         if( $n < 0x80 ) {
 245                                 continue;
 246                         } elseif( $n >= 0xf0 ) {
 247                                 $c = substr( $string, $i, 4 );
 248                                 $i += 3;
 249                         } elseif( $n >= 0xe0 ) {
 250                                 $c = substr( $string, $i, 3 );
 251                                 $i += 2;
 252                         } elseif( $n >= 0xc0 ) {
 253                                 $c = substr( $string, $i, 2 );
 254                                 $i++;
 255                         }
 256                         if( isset( $utfCheckNFC[$c] ) ) {
 257                                 # If it's NO or MAYBE, bail and do the slow check.
 258                                 return false;
 259                         }
 260                         if( isset( $utfCombiningClass[$c] ) ) {
 261                                 # Combining character? We might have to do sorting, at least.
 262                                 return false;
 263                         }
 264                 }
 265                 return true;
 266         }
 267
 268         /**
 269          * Returns true if the string is _definitely_ in NFC.
 270          * Returns false if not or uncertain.
 271          * @param string $string a UTF-8 string, altered on output to be valid UTF-8 safe for XML.
 272          */
 273         function quickIsNFCVerify( &$string ) {
 274                 # Screen out some characters that eg won't be allowed in XML
 275                 $string = preg_replace( '/[\x00-\x08\x0b\x0c\x0e-\x1f]/', UTF8_REPLACEMENT, $string );
 276
 277                 # ASCII is always valid NFC!
 278                 # If we're only ever given plain ASCII, we can avoid the overhead
 279                 # of initializing the decomposition tables by skipping out early.
 280                 if( !preg_match( '/[\x80-\xff]/', $string ) ) return true;
 281
 282                 static $checkit = null, $tailBytes = null, $utfCheckOrCombining = null;
 283                 if( !isset( $checkit ) ) {
 284                         # Load/build some scary lookup tables...
 285                         UtfNormal::loadData();
 286                         global $utfCheckNFC, $utfCombiningClass;
 287
 288                         $utfCheckOrCombining = array_merge( $utfCheckNFC, $utfCombiningClass );
 289
 290                         # Head bytes for sequences which we should do further validity checks
 291                         $checkit = array_flip( array_map( 'chr',
 292                                         array( 0xc0, 0xc1, 0xe0, 0xed, 0xef,
 293                                                    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7,
 294                                                    0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff ) ) );
 295
 296                         # Each UTF-8 head byte is followed by a certain
 297                         # number of tail bytes.
 298                         $tailBytes = array();
 299                         for( $n = 0; $n < 256; $n++ ) {
 300                                 if( $n < 0xc0 ) {
 301                                         $remaining = 0;
 302                                 } elseif( $n < 0xe0 ) {
 303                                         $remaining = 1;
 304                                 } elseif( $n < 0xf0 ) {
 305                                         $remaining = 2;
 306                                 } elseif( $n < 0xf8 ) {
 307                                         $remaining = 3;
 308                                 } elseif( $n < 0xfc ) {
 309                                         $remaining = 4;
 310                                 } elseif( $n < 0xfe ) {
 311                                         $remaining = 5;
 312                                 } else {
 313                                         $remaining = 0;
 314                                 }
 315                                 $tailBytes[chr($n)] = $remaining;
 316                         }
 317                 }
 318
 319                 # Chop the text into pure-ASCII and non-ASCII areas;
 320                 # large ASCII parts can be handled much more quickly.
 321                 # Don't chop up Unicode areas for punctuation, though,
 322                 # that wastes energy.
 323                 preg_match_all(
 324                         '/([\x00-\x7f]+|[\x80-\xff][\x00-\x40\x5b-\x5f\x7b-\xff]*)/',
 325                         $string, $matches );
 326
 327                 $looksNormal = true;
 328                 $base = 0;
 329                 $replace = array();
 330                 foreach( $matches[1] as $str ) {
 331                         $chunk = strlen( $str );
 332
 333                         if( $str{0} < "\x80" ) {
 334                                 # ASCII chunk: guaranteed to be valid UTF-8
 335                                 # and in normal form C, so skip over it.
 336                                 $base += $chunk;
 337                                 continue;
 338                         }
 339
 340                         # We'll have to examine the chunk byte by byte to ensure
 341                         # that it consists of valid UTF-8 sequences, and to see
 342                         # if any of them might not be normalized.
 343                         #
 344                         # Since PHP is not the fastest language on earth, some of
 345                         # this code is a little ugly with inner loop optimizations.
 346
 347                         $head = '';
 348                         $len = $chunk + 1; # Counting down is faster. I'm *so* sorry.
 349
 350                         for( $i = -1; --$len; ) {
 351                                 if( $remaining = $tailBytes[$c = $str{++$i}] ) {
 352                                         # UTF-8 head byte!
 353                                         $sequence = $head = $c;
 354                                         do {
 355                                                 # Look for the defined number of tail bytes...
 356                                                 if( --$len && ( $c = $str{++$i} ) >= "\x80" && $c < "\xc0" ) {
 357                                                         # Legal tail bytes are nice.
 358                                                         $sequence .= $c;
 359                                                 } else {
 360                                                         if( 0 == $len ) {
 361                                                                 # Premature end of string!
 362                                                                 # Drop a replacement character into output to
 363                                                                 # represent the invalid UTF-8 sequence.
 364                                                                 $replace[] = array( UTF8_REPLACEMENT,
 365                                                                                                         $base + $i + 1 - strlen( $sequence ),
 366                                                                                                         strlen( $sequence ) );
 367                                                                 break 2;
 368                                                         } else {
 369                                                                 # Illegal tail byte; abandon the sequence.
 370                                                                 $replace[] = array( UTF8_REPLACEMENT,
 371                                                                                                         $base + $i - strlen( $sequence ),
 372                                                                                                         strlen( $sequence ) );
 373                                                                 # Back up and reprocess this byte; it may itself
 374                                                                 # be a legal ASCII or UTF-8 sequence head.
 375                                                                 --$i;
 376                                                                 ++$len;
 377                                                                 continue 2;
 378                                                         }
 379                                                 }
 380                                         } while( --$remaining );
 381
 382                                         if( isset( $checkit[$head] ) ) {
 383                                                 # Do some more detailed validity checks, for
 384                                                 # invalid characters and illegal sequences.
 385                                                 if( $head == "\xed" ) {
 386                                                         # 0xed is relatively frequent in Korean, which
 387                                                         # abuts the surrogate area, so we're doing
 388                                                         # this check separately to speed things up.
 389
 390                                                         if( $sequence >= UTF8_SURROGATE_FIRST ) {
 391                                                                 # Surrogates are legal only in UTF-16 code.
 392                                                                 # They are totally forbidden here in UTF-8
 393                                                                 # utopia.
 394                                                                 $replace[] = array( UTF8_REPLACEMENT,
 395                                                                              $base + $i + 1 - strlen( $sequence ),
 396                                                                              strlen( $sequence ) );
 397                                                                 $head = '';
 398                                                                 continue;
 399                                                         }
 400                                                 } else {
 401                                                         # Slower, but rarer checks...
 402                                                         $n = ord( $head );
 403                                                         if(
 404                                                                 # "Overlong sequences" are those that are syntactically
 405                                                                 # correct but use more UTF-8 bytes than are necessary to
 406                                                                 # encode a character. Naïve string comparisons can be
 407                                                                 # tricked into failing to see a match for an ASCII
 408                                                                 # character, for instance, which can be a security hole
 409                                                                 # if blacklist checks are being used.
 410                                                                ($n  < 0xc2 && $sequence <= UTF8_OVERLONG_A)
 411                                                                 || ($n == 0xe0 && $sequence <= UTF8_OVERLONG_B)
 412                                                                 || ($n == 0xf0 && $sequence <= UTF8_OVERLONG_C)
 413
 414                                                                 # U+FFFE and U+FFFF are explicitly forbidden in Unicode.
 415                                                                 || ($n == 0xef &&
 416                                                                            ($sequence == UTF8_FFFE)
 417                                                                         || ($sequence == UTF8_FFFF) )
 418
 419                                                                 # Unicode has been limited to 21 bits; longer
 420                                                                 # sequences are not allowed.
 421                                                                 || ($n >= 0xf0 && $sequence > UTF8_MAX) ) {
 422
 423                                                                 $replace[] = array( UTF8_REPLACEMENT,
 424                                                                                     $base + $i + 1 - strlen( $sequence ),
 425                                                                                     strlen( $sequence ) );
 426                                                                 $head = '';
 427                                                                 continue;
 428                                                         }
 429                                                 }
 430                                         }
 431
 432                                         if( isset( $utfCheckOrCombining[$sequence] ) ) {
 433                                                 # If it's NO or MAYBE, we'll have to rip
 434                                                 # the string apart and put it back together.
 435                                                 # That's going to be mighty slow.
 436                                                 $looksNormal = false;
 437                                         }
 438
 439                                         # The sequence is legal!
 440                                         $head = '';
 441                                 } elseif( $c < "\x80" ) {
 442                                         # ASCII byte.
 443                                         $head = '';
 444                                 } elseif( $c < "\xc0" ) {
 445                                         # Illegal tail bytes
 446                                         if( $head == '' ) {
 447                                                 # Out of the blue!
 448                                                 $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 449                                         } else {
 450                                                 # Don't add if we're continuing a broken sequence;
 451                                                 # we already put a replacement character when we looked
 452                                                 # at the broken sequence.
 453                                                 $replace[] = array( '', $base + $i, 1 );
 454                                         }
 455                                 } else {
 456                                         # Miscellaneous freaks.
 457                                         $replace[] = array( UTF8_REPLACEMENT, $base + $i, 1 );
 458                                         $head = '';
 459                                 }
 460                         }
 461                         $base += $chunk;
 462                 }
 463                 if( count( $replace ) ) {
 464                         # There were illegal UTF-8 sequences we need to fix up.
 465                         $out = '';
 466                         $last = 0;
 467                         foreach( $replace as $rep ) {
 468                                 list( $replacement, $start, $length ) = $rep;
 469                                 if( $last < $start ) {
 470                                         $out .= substr( $string, $last, $start - $last );
 471                                 }
 472                                 $out .= $replacement;
 473                                 $last = $start + $length;
 474                         }
 475                         if( $last < strlen( $string ) ) {
 476                                 $out .= substr( $string, $last );
 477                         }
 478                         $string = $out;
 479                 }
 480                 return $looksNormal;
 481         }
 482
 483         # These take a string and run the normalization on them, without
 484         # checking for validity or any optimization etc. Input must be
 485         # VALID UTF-8!
 486         /**
 487          * @param string $string
 488          * @return string
 489          * @access private
 490          */
 491         function NFC( $string ) {
 492                 return UtfNormal::fastCompose( UtfNormal::NFD( $string ) );
 493         }
 494
 495         /**
 496          * @param string $string
 497          * @return string
 498          * @access private
 499          */
 500         function NFD( $string ) {
 501                 UtfNormal::loadData();
 502                 global $utfCanonicalDecomp;
 503                 return UtfNormal::fastCombiningSort(
 504                         UtfNormal::fastDecompose( $string, $utfCanonicalDecomp ) );
 505         }
 506
 507         /**
 508          * @param string $string
 509          * @return string
 510          * @access private
 511          */
 512         function NFKC( $string ) {
 513                 return UtfNormal::fastCompose( UtfNormal::NFKD( $string ) );
 514         }
 515
 516         /**
 517          * @param string $string
 518          * @return string
 519          * @access private
 520          */
 521         function NFKD( $string ) {
 522                 global $utfCompatibilityDecomp;
 523                 if( !isset( $utfCompatibilityDecomp ) ) {
 524                         require_once( 'UtfNormalDataK.inc' );
 525                 }
 526                 return UtfNormal::fastCombiningSort(
 527                         UtfNormal::fastDecompose( $string, $utfCompatibilityDecomp ) );
 528         }
 529
 530
 531         /**
 532          * Perform decomposition of a UTF-8 string into either D or KD form
 533          * (depending on which decomposition map is passed to us).
 534          * Input is assumed to be *valid* UTF-8. Invalid code will break.
 535          * @access private
 536          * @param string $string Valid UTF-8 string
 537          * @param array $map hash of expanded decomposition map
 538          * @return string a UTF-8 string decomposed, not yet normalized (needs sorting)
 539          */
 540         function fastDecompose( $string, &$map ) {
 541                 UtfNormal::loadData();
 542                 $len = strlen( $string );
 543                 $out = '';
 544                 for( $i = 0; $i < $len; $i++ ) {
 545                         $c = $string{$i};
 546                         $n = ord( $c );
 547                         if( $n < 0x80 ) {
 548                                 # ASCII chars never decompose
 549                                 # THEY ARE IMMORTAL
 550                                 $out .= $c;
 551                                 continue;
 552                         } elseif( $n >= 0xf0 ) {
 553                                 $c = substr( $string, $i, 4 );
 554                                 $i += 3;
 555                         } elseif( $n >= 0xe0 ) {
 556                                 $c = substr( $string, $i, 3 );
 557                                 $i += 2;
 558                         } elseif( $n >= 0xc0 ) {
 559                                 $c = substr( $string, $i, 2 );
 560                                 $i++;
 561                         }
 562                         if( isset( $map[$c] ) ) {
 563                                 $out .= $map[$c];
 564                                 continue;
 565                         } else {
 566                                 if( $c >= UTF8_HANGUL_FIRST && $c <= UTF8_HANGUL_LAST ) {
 567                                         # Decompose a hangul syllable into jamo;
 568                                         # hardcoded for three-byte UTF-8 sequence.
 569                                         # A lookup table would be slightly faster,
 570                                         # but adds a lot of memory & disk needs.
 571                                         #
 572                                         $index = ( (ord( $c{0} ) & 0x0f) << 12
 573                                                  | (ord( $c{1} ) & 0x3f) <<  6
 574                                                  | (ord( $c{2} ) & 0x3f) )
 575                                                - UNICODE_HANGUL_FIRST;
 576                                         $l = IntVal( $index / UNICODE_HANGUL_NCOUNT );
 577                                         $v = IntVal( ($index % UNICODE_HANGUL_NCOUNT) / UNICODE_HANGUL_TCOUNT);
 578                                         $t = $index % UNICODE_HANGUL_TCOUNT;
 579                                         $out .= "\xe1\x84" . chr( 0x80 + $l ) . "\xe1\x85" . chr( 0xa1 + $v );
 580                                         if( $t >= 25 ) {
 581                                                 $out .= "\xe1\x87" . chr( 0x80 + $t - 25 );
 582                                         } elseif( $t ) {
 583                                                 $out .= "\xe1\x86" . chr( 0xa7 + $t );
 584                                         }
 585                                         continue;
 586                                 }
 587                         }
 588                         $out .= $c;
 589                 }
 590                 return $out;
 591         }
 592
 593         /**
 594          * Sorts combining characters into canonical order. This is the
 595          * final step in creating decomposed normal forms D and KD.
 596          * @access private
 597          * @param string $string a valid, decomposed UTF-8 string. Input is not validated.
 598          * @return string a UTF-8 string with combining characters sorted in canonical order
 599          */
 600         function fastCombiningSort( $string ) {
 601                 UtfNormal::loadData();
 602                 global $utfCombiningClass;
 603                 $len = strlen( $string );
 604                 $out = '';
 605                 $combiners = array();
 606                 $lastClass = -1;
 607                 for( $i = 0; $i < $len; $i++ ) {
 608                         $c = $string{$i};
 609                         $n = ord( $c );
 610                         if( $n >= 0x80 ) {
 611                                 if( $n >= 0xf0 ) {
 612                                         $c = substr( $string, $i, 4 );
 613                                         $i += 3;
 614                                 } elseif( $n >= 0xe0 ) {
 615                                         $c = substr( $string, $i, 3 );
 616                                         $i += 2;
 617                                 } elseif( $n >= 0xc0 ) {
 618                                         $c = substr( $string, $i, 2 );
 619                                         $i++;
 620                                 }
 621                                 if( isset( $utfCombiningClass[$c] ) ) {
 622                                         $lastClass = $utfCombiningClass[$c];
 623                                         @$combiners[$lastClass] .= $c;
 624                                         continue;
 625                                 }
 626                         }
 627                         if( $lastClass ) {
 628                                 ksort( $combiners );
 629                                 $out .= implode( '', $combiners );
 630                                 $combiners = array();
 631                         }
 632                         $out .= $c;
 633                         $lastClass = 0;
 634                 }
 635                 if( $lastClass ) {
 636                         ksort( $combiners );
 637                         $out .= implode( '', $combiners );
 638                 }
 639                 return $out;
 640         }
 641
 642         /**
 643          * Produces canonically composed sequences, i.e. normal form C or KC.
 644          *
 645          * @access private
 646          * @param string $string a valid UTF-8 string in sorted normal form D or KD. Input is not validated.
 647          * @return string a UTF-8 string with canonical precomposed characters used where possible
 648          */
 649         function fastCompose( $string ) {
 650                 UtfNormal::loadData();
 651                 global $utfCanonicalComp, $utfCombiningClass;
 652                 $len = strlen( $string );
 653                 $out = '';
 654                 $lastClass = -1;
 655                 $lastHangul = 0;
 656                 $startChar = '';
 657                 $combining = '';
 658                 $x1 = ord(substr(UTF8_HANGUL_VBASE,0,1));
 659                 $x2 = ord(substr(UTF8_HANGUL_TEND,0,1));
 660                 for( $i = 0; $i < $len; $i++ ) {
 661                         $c = $string{$i};
 662                         $n = ord( $c );
 663                         if( $n < 0x80 ) {
 664                                 # No combining characters here...
 665                                 $out .= $startChar;
 666                                 $out .= $combining;
 667                                 $startChar = $c;
 668                                 $combining = '';
 669                                 $lastClass = 0;
 670                                 continue;
 671                         } elseif( $n >= 0xf0 ) {
 672                                 $c = substr( $string, $i, 4 );
 673                                 $i += 3;
 674                         } elseif( $n >= 0xe0 ) {
 675                                 $c = substr( $string, $i, 3 );
 676                                 $i += 2;
 677                         } elseif( $n >= 0xc0 ) {
 678                                 $c = substr( $string, $i, 2 );
 679                                 $i++;
 680                         }
 681                         $pair = $startChar . $c;
 682                         if( $n > 0x80 ) {
 683                                 if( isset( $utfCombiningClass[$c] ) ) {
 684                                         # A combining char; see what we can do with it
 685                                         $class = $utfCombiningClass[$c];
 686                                         if( !empty( $startChar ) &&
 687                                                 $lastClass < $class &&
 688                                                 $class > 0 &&
 689                                                 isset( $utfCanonicalComp[$pair] ) ) {
 690                                                 $startChar = $utfCanonicalComp[$pair];
 691                                                 $class = 0;
 692                                         } else {
 693                                                 $combining .= $c;
 694                                         }
 695                                         $lastClass = $class;
 696                                         $lastHangul = 0;
 697                                         continue;
 698                                 }
 699                         }
 700                         # New start char
 701                         if( $lastClass == 0 ) {
 702                                 if( isset( $utfCanonicalComp[$pair] ) ) {
 703                                         $startChar = $utfCanonicalComp[$pair];
 704                                         $lastHangul = 0;
 705                                         continue;
 706                                 }
 707                                 if( $n >= $x1 && $n <= $x2 ) {
 708                                         # WARNING: Hangul code is painfully slow.
 709                                         # I apologize for this ugly, ugly code; however
 710                                         # performance is even more teh suck if we call
 711                                         # out to nice clean functions. Lookup tables are
 712                                         # marginally faster, but require a lot of space.
 713                                         #
 714                                         if( $c >= UTF8_HANGUL_VBASE &&
 715                                                 $c <= UTF8_HANGUL_VEND &&
 716                                                 $startChar >= UTF8_HANGUL_LBASE &&
 717                                                 $startChar <= UTF8_HANGUL_LEND ) {
 718                                                 #
 719                                                 #$lIndex = utf8ToCodepoint( $startChar ) - UNICODE_HANGUL_LBASE;
 720                                                 #$vIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_VBASE;
 721                                                 $lIndex = ord( $startChar{2} ) - 0x80;
 722                                                 $vIndex = ord( $c{2}         ) - 0xa1;
 723
 724                                                 $hangulPoint = UNICODE_HANGUL_FIRST +
 725                                                         UNICODE_HANGUL_TCOUNT *
 726                                                         (UNICODE_HANGUL_VCOUNT * $lIndex + $vIndex);
 727
 728                                                 # Hardcode the limited-range UTF-8 conversion:
 729                                                 $startChar = chr( $hangulPoint >> 12 & 0x0f | 0xe0 ) .
 730                                                                          chr( $hangulPoint >>  6 & 0x3f | 0x80 ) .
 731                                                                          chr( $hangulPoint       & 0x3f | 0x80 );
 732                                                 $lastHangul = 0;
 733                                                 continue;
 734                                         } elseif( $c >= UTF8_HANGUL_TBASE &&
 735                                                           $c <= UTF8_HANGUL_TEND &&
 736                                                           $startChar >= UTF8_HANGUL_FIRST &&
 737                                                           $startChar <= UTF8_HANGUL_LAST &&
 738                                                           !$lastHangul ) {
 739                                                 # $tIndex = utf8ToCodepoint( $c ) - UNICODE_HANGUL_TBASE;
 740                                                 $tIndex = ord( $c{2} ) - 0xa7;
 741                                                 if( $tIndex < 0 ) $tIndex = ord( $c{2} ) - 0x80 + (0x11c0 - 0x11a7);
 742
 743                                                 # Increment the code point by $tIndex, without
 744                                                 # the function overhead of decoding and recoding UTF-8
 745                                                 #
 746                                                 $tail = ord( $startChar{2} ) + $tIndex;
 747                                                 if( $tail > 0xbf ) {
 748                                                         $tail -= 0x40;
 749                                                         $mid = ord( $startChar{1} ) + 1;
 750                                                         if( $mid > 0xbf ) {
 751                                                                 $startChar{0} = chr( ord( $startChar{0} ) + 1 );
 752                                                                 $mid -= 0x40;
 753                                                         }
 754                                                         $startChar{1} = chr( $mid );
 755                                                 }
 756                                                 $startChar{2} = chr( $tail );
 757
 758                                                 # If there's another jamo char after this, *don't* try to merge it.
 759                                                 $lastHangul = 1;
 760                                                 continue;
 761                                         }
 762                                 }
 763                         }
 764                         $out .= $startChar;
 765                         $out .= $combining;
 766                         $startChar = $c;
 767                         $combining = '';
 768                         $lastClass = 0;
 769                         $lastHangul = 0;
 770                 }
 771                 $out .= $startChar . $combining;
 772                 return $out;
 773         }
 774
 775         /**
 776          * This is just used for the benchmark, comparing how long it takes to
 777          * interate through a string without really doing anything of substance.
 778          * @param string $string
 779          * @return string
 780          */
 781         function placebo( $string ) {
 782                 $len = strlen( $string );
 783                 $out = '';
 784                 for( $i = 0; $i < $len; $i++ ) {
 785                         $out .= $string{$i};
 786                 }
 787                 return $out;
 788         }
 789 }
 790
 791 ?>